Date of submission: 12
Download 0.72 Mb.
|
- Navigate this page:
- 6.4 Creation of a change map for 1988-2010
- 6.5 Validation of change map
6.3 Austin population changeAustin has experienced a growing population and the population figures from the US census bureau suggest that the population of Austin grew at its fastest rate in the late 1990’s. Figure 43: Austin population figures from 1990 – 2010. Source, US Census Bureau 2011 This links with the change detection maps in that the greatest amount of urban growth appeared in the second epoch, from 1995 – 2003. The steady rise in population in Austin throughout the duration of the study reflects the strengthening local economy and job opportunities in the area. There is acknowledged to be a relationship between population change and urban growth (Masek et al, 2000) and this also appears to hold true for Austin from the period 1988 – 2010. 6.4 Creation of a change map for 1988-2010After the detailed change analysis had taken place the next step was to use the Idrisi LCM to create a decrease in NDVI prediction map for the third epoch, from 2003-2010, which would be added to the change in the first two epochs to create a composite map from 1988 – 2010. This could then be compared against the actual composite decrease in NDVI map for 1988 – 2010 and validated. The methods as described in sections 5.6 and 5.7 were employed to create a transition potential and the change prediction tab was then selected. First the prediction year was input into the change demand modelling panel and the Markov chain was selected to model the transition. 
Figure 44: Change demand modelling dropdown in Change Prediction tab, LCM The change allocation panel was then populated, including the frequency with which dynamic explanatory variables were to be recalculated. The distance from sprawl was a dynamic variable as this will change over time. Given that sprawl can occur rapidly, three recalculation stages were selected, which means this variable would be updated three times over the seven years from 2003 – 2010, that the model was predicting, once every 2.33 years. This does take longer to run the model, as at each stage the explanatory variables are being resubmitted to the MLP, which applies the originally calculated weights to the revised explanatory variables to calculate new transition potentials. Once the model had completed, a land cover model for the third epoch, from 2003 – 2010 was generated. This was added to the previous changes from the first two epochs and is as below, alongside the actual change map for all three epochs. Hard prediction Actual change map  
Figure 45: Comparison of actual decrease in NDVI map for three epochs and prediction created in LCM 6.5 Validation of change mapA visual inspection shows that both maps are reasonably similar to each other. However, there are noticeable differences between them. The actual map shows greater scatter of sprawl across the Austin area (for example in the west and north of Austin), whereas the prediction failed to identify this. The predicted map tends to overestimate the amount of sprawl in the north east of the study area. In order to explore some of these patterns further there are two options within Idrisi. The first is using the VALIDATE module. This allows the user to compare the actual and predicted map and it returns an agreement for the entire map. This module was run, with the results generated as below: 
Figure 46: Validation results This suggests a very close match between the actual and predicted map, however this is because it evaluates agreement for the entire map, and not just for the class transitions which we are interested in. This is a naïve method and the very high agreement between maps is only present due to a large signal of persistence on the landscape (Pontius Jr and Chen, 2006). The LCM provides a more robust validation option, which runs a three way cross tabulation between the later land cover map in the input, the prediction map that was created and the map of reality and this examines the agreement between two maps that show the same categorical variable. This model gives an indication of the accuracy of the predicted map and it splits this into three categories: Hits (green) – Model predicted change and it changed Misses (red) – Model predicted persistence and it changed False Alarms (yellow) – Model predicted change and it persisted. This module was run and the map below was generated.
In terms of actual urban growth, the actual composite epoch 3 map displayed 328 km2 of growth. The predicted map gave 333 km2 which is only a 5km2 difference, so whilst location of growth was not always accurate, the actual amount was very close. Download 0.72 Mb. Share with your friends:
|
